From 99a94a30467956cd2264996e7498cbc43f83ef64 Mon Sep 17 00:00:00 2001
From: foefl <f.foerster@d-opt.com>
Date: Thu, 12 Mar 2026 12:26:45 +0100
Subject: [PATCH] add automatic build and publishing scripts

---
 .gitignore                                  |  1 +
 build_llama-cpp-python.bat                  | 66 +++++++++++++++++++++
 pdm.lock                                    | 13 +++-
 pyproject.toml                              |  2 +-
 scripts/publish.ps1                         |  2 +-
 src/{py311 => llama_cpp_wheels}/__init__.py |  0
 6 files changed, 81 insertions(+), 3 deletions(-)
 create mode 100644 build_llama-cpp-python.bat
 rename src/{py311 => llama_cpp_wheels}/__init__.py (100%)

diff --git a/.gitignore b/.gitignore
index 21e6a91..549accc 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,4 +1,5 @@
 # own
+llama_cpp_python_src/
 prototypes/
 data/
 reports/
diff --git a/build_llama-cpp-python.bat b/build_llama-cpp-python.bat
new file mode 100644
index 0000000..f8556d0
--- /dev/null
+++ b/build_llama-cpp-python.bat
@@ -0,0 +1,66 @@
+@echo off
+SETLOCAL EnableDelayedExpansion
+
+:: path to VS Build Tools (example for VS 2022 Community) (otherwise not found in default terminal)
+set "VS_PATH=C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Auxiliary\Build\vcvars64.bat"
+
+if exist "%VS_PATH%" (
+    call "%VS_PATH%"
+) else (
+    echo [X] Setup script for Visual Studio env variables could not be found!
+    pause
+    exit /b
+)
+
+:: --- Configuration ---
+set "REPO_URL=https://github.com/abetlen/llama-cpp-python.git"
+set "SOURCE_DIR=llama_cpp_python_src"
+set "WHEEL_DIR=.\dist"
+
+:: Optional: specify CUDA architecture (e.g. "89" für RTX 1000 Ada Laptop GPU)
+:: "native" architecture only possible if device is identified properly
+:: see https://developer.nvidia.com/cuda/gpus for compute compatibility
+:: set "CMAKE_ARGS=-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=native"
+set "CMAKE_ARGS=-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES='89'"
+set "FORCE_CMAKE=1"
+set "CMAKE_BUILD_PARALLEL_LEVEL=4"
+set "MAX_JOBS=4"
+
+echo [*] Check source directory...
+
+if exist "%SOURCE_DIR%" (
+    echo [*] Directory exists. Update repository...
+    cd "%SOURCE_DIR%"
+    git fetch --all
+    git reset --hard origin/main
+    git pull origin main
+    :: update submodules (contains the actual llama.cpp)
+    git submodule update --init --recursive
+) else (
+    echo [*] Clone repository...
+    git clone --recurse-submodules %REPO_URL% %SOURCE_DIR%
+    cd "%SOURCE_DIR%"
+)
+
+echo [*] Start compilation und wheel generation...
+
+:: build wheel from current directory (.)
+@REM pdm run -p ..\ python --version
+@REM pdm run -p ..\ python -m pip --version
+@REM pause
+@REM exit /b
+
+pdm run -p ..\ python -m pip wheel . --wheel-dir="..\%WHEEL_DIR%" --no-cache-dir --no-deps -vv
+
+if %ERRORLEVEL% EQU 0 (
+    echo.
+    echo [!] SUCCESS: Wheel with CUDA support placed in: %WHEEL_DIR%
+    dir "..\%WHEEL_DIR%\llama_cpp_python*.whl"
+) else (
+    echo.
+    echo [X] ERROR: Build process failed.
+)
+
+cd ..
+pause
+ENDLOCAL
\ No newline at end of file
diff --git a/pdm.lock b/pdm.lock
index f96da8b..8a411b4 100644
--- a/pdm.lock
+++ b/pdm.lock
@@ -5,7 +5,7 @@
 groups = ["default", "dev", "lint"]
 strategy = ["inherit_metadata"]
 lock_version = "4.5.0"
-content_hash = "sha256:b23e8c0ac043347792e09d8cc5ed2b185120ad3cec1a2a179842a690a9618d33"
+content_hash = "sha256:281a45dbbe9fe4927e83d52f27e165b12498964c56e3d2600e5a5562ba627a23"
 
 [[metadata.targets]]
 requires_python = ">=3.11"
@@ -423,6 +423,17 @@ files = [
     {file = "pdoc3-0.11.6.tar.gz", hash = "sha256:1ea5e84b87a754d191fb64bf5e517ca6c50d0d84a614c1efecf6b46d290ae387"},
 ]
 
+[[package]]
+name = "pip"
+version = "26.0.1"
+requires_python = ">=3.9"
+summary = "The PyPA recommended tool for installing Python packages."
+groups = ["default"]
+files = [
+    {file = "pip-26.0.1-py3-none-any.whl", hash = "sha256:bdb1b08f4274833d62c1aa29e20907365a2ceb950410df15fc9521bad440122b"},
+    {file = "pip-26.0.1.tar.gz", hash = "sha256:c4037d8a277c89b320abe636d59f91e6d0922d08a05b60e85e53b296613346d8"},
+]
+
 [[package]]
 name = "platformdirs"
 version = "4.9.4"
diff --git a/pyproject.toml b/pyproject.toml
index 7a74ce7..fde83ae 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -5,7 +5,7 @@ description = "project to build CUDA-enabled builds of Python bindings for 'llam
 authors = [
     {name = "Florian Förster", email = "f.foerster@d-opt.com"},
 ]
-dependencies = []
+dependencies = ["pip>=26.0.1"]
 requires-python = ">=3.11"
 readme = "README.md"
 license = {text = "LicenseRef-Proprietary"}
diff --git a/scripts/publish.ps1 b/scripts/publish.ps1
index d16fc9c..e1b2f7f 100644
--- a/scripts/publish.ps1
+++ b/scripts/publish.ps1
@@ -1 +1 @@
-pdm publish -r local --skip-existing
\ No newline at end of file
+pdm publish -r local -d .\dist --skip-existing --no-build
\ No newline at end of file
diff --git a/src/py311/__init__.py b/src/llama_cpp_wheels/__init__.py
similarity index 100%
rename from src/py311/__init__.py
rename to src/llama_cpp_wheels/__init__.py